Algorithms for Differentially Private Multi-Armed Bandits

TL;DR

This paper introduces differentially private algorithms for stochastic multi-armed bandits that achieve optimal regret bounds, improving upon previous methods through a novel interval-based mechanism and validated by experiments.

Contribution

It presents the first differentially private UCB algorithms with optimal regret bounds, utilizing a new interval-based mechanism for enhanced privacy-utility trade-offs.

Findings

Achieved $(\epsilon,\delta)$-DP algorithms with $O(rac{1}{\epsilon} + \log T)$ regret.

Significant improvement over previous poly-log regret bounds.

Experimental results confirm theoretical guarantees.

Abstract

We present differentially private algorithms for the stochastic Multi-Armed Bandit (MAB) problem. This is a problem for applications such as adaptive clinical trials, experiment design, and user-targeted advertising where private information is connected to individual rewards. Our major contribution is to show that there exist $(\epsilon, \delta)$ differentially private variants of Upper Confidence Bound algorithms which have optimal regret, $O(\epsilon^{-1} + \log T)$. This is a significant improvement over previous results, which only achieve poly-log regret $O(\epsilon^{-2} \log^{2} T)$, because of our use of a novel interval-based mechanism. We also substantially improve the bounds of previous family of algorithms which use a continual release mechanism. Experiments clearly validate our theoretical bounds.